Stephan Klasen

Report
The Multidimensional Poverty Index:
Achievements, Conceptual, and Empirical
Issues
Caroline Dotter
Stephan Klasen
Universität Göttingen
Milorad Kovacevic
HDRO
HDRO Workshop
March 4, 2013
1
The MPI
• Measuring acute multidimensional poverty;
• Based on dual cut-off approach (1/3);
• Dimensions: Health (mortality and nutrition),
Education (years and enrolement), Standard of
living (house, water, sanitation, electricity, cook
fuel, assets);
• MPI = Headcount * Intensity;
• Data used: DHS, MICS, WHS
• Calculated for some 110 countries (increasingly
available for more than 1 period);
2
In praise of an MPI-type Indicator
• Direct multidimensional complement/competitor to $ a
day indicator;
– Similar breadth and coverage
– Could possibly calculate and monitor global poverty;
• Also based on capability approach (as is the HDI);
• Actionable and policy-relevant at the national (and subnational level); advantage largely unexploited by UNDP;
• Consistent with reasonable set of poverty measurement
axioms (in contrast to HPI);
• Based on high quality and comparable data, with
potential to measure poverty over time;
3
Conceptual Issues
• Dual cut-off navigates between union and intersection
approach
– But leads to formal and interpretational problems: deprivations
entirely ignored below the cut-off seems problematic;
– Union approach conceptually to be preferred?
• Neglect of inequality in the spread of dimensions across
the population, which is also problematic;
– Proposal by Rippin: In the poverty identification step, use square
of weighted deprivation share as poverety indicator (and add
those up in aggregation step);
– Other proposals in the literature;
• Use of intensity in the MPI:
– cannot compare with $ a day headcount
– little variation in intensity (heavily driven by second cut-off);
– use headcount as headline indicator with intensity-inequality
sensitive measure as complementary indicator?
4
Empirical Issues
• WHS limiting and problematic (and now superfluous?);
suggestion to just use MICS and DHS;
• Standard of living:
– Unclear interpretation of electricity access (unequal use!),
cooking fuel (depends on cooking situation), and sanitation
(needs differ across rural/urban, regions);
– Quite large influence on overall MPI;
– 3 indicators would suffice (and capture others as well): floor,
assets, and drinking water;
• Enrolments:
– One child not enrolled, household deprived;
– Problem of late enrolments;
– Adjust time window to allow for late enrolments (e.g. allow for 2
years late enrolment);
5
Share of population deprived in enrolment
Whole population
Original enrolment window 25.32
Shorter enrolment window 17.42
Population with school-aged children
(original category)
38.87
26.71
6
Empirical Issues
• Mortality:
– Only consider recent child deaths (MICS: only consider deaths of
women who gave births in last 10 years?);
• Nutrition:
– BMI of adults and childhood undernutrition cut-offs not directly
comparable;
– BMI and underweight subject to bias due to nutrition transition;
– Focus on children beyond 6 months?
– Proposal: Just focus on childhood undernutrition and stunting;
• Education:
– Cut-off (one person with 5 years enough for non-deprivation) and
implies perfect economies of scale (asymmetry);
– Proposal: deprived if less than 50% of adults have 5 years+
7
Empirical Issues
• Asymmetric cut-offs in health, enrolment, nutrition,
education:
– Has systematic influence on impact of household size on MPI;
– Not clear that asymmetries are justified;
– Define cut-offs with respect to hh size (e.g. 20% of children are
undernourished);
• Ineligible population:
– No children (in school-going age or with nutritional
measurement);
– Presumed non-deprived in MPI (serious problem and bias!);
– Makes severe poverty near-impossible for hh without eligible
population;
– A serious problem of differential importance across countries;
8
Relative importance of households without eligible population
base
Nutrition (health)
Mortality (health)
all
9.1%
17.84%
Armenia
14.81%
23.58%
India
8.57%
17.13%
Ethiopia
11.07%
21.23%
Old hh (above35)
28.44%
32.48%
Enrollment (education)
36.97%
51.25%
37.90%
24.38%
38.24%
• All solutions problematic:
•Non-deprivation assumption;
•Dropping observations;
•Using other indicator from same dimension;
•Proposal: Hybrid approach: Use indicator from same dimension if
one indicator is missing, and adjust overall MPI cut-off if both are
missing (can be easily implemented);
•Advantage: Keeps all observations in, uses information to
maximum extent; likely to generate least bias;
•Disadvantage: Decompositoion no longer possible;
9
Implementing the Proposals
• A reduced and (more robust) MPI?
–
–
–
–
–
3 standard of living indicators;
Nutrition: stunting (>6mts)
Mortality: only recent deaths;
Enrolment: allow for late enrolment;
Cut-offs more uniform (>20% affected in nutrition, enrolment,
mortality, <50% with 5 years+ education);
– Hybrid approach for ineligible population;
• Implement approach using DHS for Armenia, Ethiopia,
and India;
• Changes incidence (mainly due to education cut-off), but
also correlates of poverty (e.g. hh size);
10
Table 2: Multidimensional Poverty across sub-groups and countries
H
A
All
54.85%
55.28%
Urban
20.82%
48.47%
Rural
69.44%
56.15%
small hh
44.31%
49.31%
medium hh
50.46%
54.73%
large hh
66.03%
57.63%
female-headed hh
54.68%
56.51%
above 35
51.59%
55.78%
below 35
55.81%
55.14%
Armenia
0.57%
38.24%
Ethiopia
90.48%
64.59%
India
52.76%
53.17%
Improved multidimensional poverty estimation
H
all
urban
rural
small hh
medium hh
large hh
female-headed hh
above 35
below 35
Armenia
Ethiopia
India
MPI
0.303206
0.100921
0.389922
0.218485
0.276131
0.380515
0.308982
0.287795
0.307708
0.002194
0.584382
0.28055
A
60.28%
27.22%
73.24%
53.53%
57.58%
64.78%
59.98%
57.44%
60.02%
2.96%
92.25%
57.82%
MPI
61.46%
55.57%
62.89%
59.77%
61.70%
62.89%
61.41%
60.77%
62.20%
46.89%
69.25%
60.37%
0.370522
0.151271
0.460657
0.319907
0.355257
0.407391
0.368327
0.349068
0.373302
0.013863
0.638847
0.349068
11
Conclusion
• MPI has been a good start to develop internationally
comparable multidimensional poverty indicator;
• But there are open issues and problems, and
refinements at conceptual and empirical level warranted
• Conceptual level: Union approach, incorporating
inequality, headcount the headline indicator?
• Empirical level: Changes to indicators, cut-offs, data sets
used, and assumptions about ineligible population;
• Most issues can be readily addressed and are worth
addressing.
12
Original
(current) MPI
Headline index
Complementary
indicators of
poverty
Cut-off
approach
Dimension cutoff
Dimension
weights
Within
dimension
weights
MPI
Headcount,
Intensity
Dual
Absolute
New proposal
Implications
Headcount of MP
Better comparability
with income poverty
Intensity, Inequality
Intensity of MP; but
Which approach to
inequality of
deprivation ?
Dual → MP
Union approach →
Measure of
deprivation, inequality
in deprivation
Consider ‘relative’
cut-offs
Equal (1/3)
Equal (1/3)
Equal
Equal
Possible differentiation
of deprivation and
multidimensional
poverty. More analytic
power.
Hard to implement and
also arbitrary?
13
Living
standard
Education
Health
Original
(current) MPI
Drinking water,
sanitation,
electricity,
cooking fuel,
floor,
assets
Any schoolEnrollment aged child is not
(ages 6-14) attending school
in grades 1 to 8
Years of
schooling is a
Years of
public good
schooling
(no one has 5 or
(age 15 and
more years of
above)
primary
education)
• BMI for adults
Nutrition
• Weight-for-age
for children
Death of
children any
Mortality
age, no
reference period
New proposal
Drinking water,
floor,
assets
Implications
Reduces the
importance of living
standard;
Reduces the
headcount
Shorter the enrollment
window by 2 years (8 to Reduces the
14); size adjustment (1 headcount
in 5)
Some economies of
scale but not full;
Size adjustment (1 in 2
adults)
Increases the
headcount
• Exclude BMI for adults No health indicator
• Height-for-age
for adults; reduces
for children
the headcount
Death of children below
age 5 in the past 5
Reference period ?
14
years;
Original
(current) MPI
No eligible
population
Severe
poverty
Enrollment,
HH is nonHealth
deprived
New proposal
Hybrid approach:
1.Double the weight
on adult education
2.BMI of adults
3.Lower cut-off: 2/9
Implications
Large number of hh
(20%); messy
calculation
At least 50% of
Deprived in
eligible population in
more than 1/2 of HH is deprived in
Reduced headcount
weighted
enrollment and health;
indicators
no assets;
Cut-off 1/3
15

similar documents